NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/benchmar/1703

Internet Message Format | 1992-11-15 | 1.7 KB

Xref: sparky comp.benchmarks:1703 comp.arch:10788
Path: sparky!uunet!bcstec!silverm
From: [e-mail] (Jeff Silverman)
Newsgroups: comp.benchmarks,comp.arch
Subject: Re: DEC ALPHA Performance Claims
Message-ID: <[e-mail]>
Date: 16 Nov 92 06:36:50 GMT
References: <[e-mail]>
Followup-To: comp.benchmarks
Organization: Boeing Computer Services, Seattle
Lines: 24

There is a lot of discussion on this thread about hardware vs. software speed
up.  I suspect that the question is moot.  One of the features, attributes,
whatever of the Alpha is that the compiler is supposed to sort instructions
so that the hardware pipeline works optimally.  This means that, for all
intents and purposes, the compiler and the CPU are closely coupled in a
new and unique way.  It also has some rather interesting side effects:

1) Newer implementations of the alpha may need or take advantage of new ways
of ordering instructions.  This in turn means that different alpha
implementations may need or take advantage of different executables or object
modules, each compiled for its own processor.  Or maybe the processors will
take a performance hit in the name of portability.  Who can say?

2) It will be interesting to see if third party vendors come out with language
processors (I.e. compilers) for the Alpha, if the compiler and the CPUs are
so intimately familiar with one another.

I don't know enough about the Alpha to answer these questions - and it may
be the case that the questions won't be answerable until somebody tries.
One test will be to measure how long it takes for somebody to announce an
Ada for the Alpha.  We'll see.

Jeff Silverman, Boeing.